Revolutionizing Referrals A Case Study: Accelerating Patient Admittance with Generative AI

Indu S
October 25, 2024

Introduction:

Imagine you're feeling unwell and go to see your family doctor. After examining you, your doctor might decide that you need to see a specialist, like a heart doctor or a skin doctor. This is called a patient referral. A primary care physician initiates a referral to ensure that the patient receives important and timely recommendations from specialists. In order for a successful and timely referral process, the referring and referred partners has to exchange patient details without causing an undue burden for clinicians or patients.

Referral management is a core activity of care coordination, valued by all stakeholders. Optimizing the referrals process can reduce the burden on staff caused by missed appointments and incomplete patient information. During referral requests a lot of patient information and documentation like prior treatments, related imaging or lab results are shared between the PCP (primary care physician) and Specialist.

One of the primary challenges faced by clinical coordinators in referral management is the verification and validation of patient referral information and timely decision on patient admittance. Here we are presenting a Generative AI powered solution implemented for a client where we created an impact using gen-AI models to accelerate the patient admittance process.

Problem Statement:

In many cases, the patient health record data that are received from various third-party sources are in heterogeneous format including text, images, lab reports, and other medical documentation. Each facility follows a unique patient admittance procedure based on the responses to its individual intake questionnaire. Clinical coordinators invest significant time in reading referral details and responding to associated questions. This comprehensive manual review process leads to clinician burnout and delay in patient admittance.

Pain Points:

1. Patient data are in heterogeneous format.

2. Care transitions between facilities are
often slow. Many facilities rely on unique manual processes that take hours-or even days to process. Studies reveal that many facilities face delays of four to five hours or more for admitting patients.

3. The document review for patient admittance is a manual process that leads to clinician burnout.

Solution:

Objectives:

1. Develop an AI-powered solution to automate Referral Question and answering process using Generative AI, thus enabling quick referral admission decisions.

2. Enhance and optimize the overall workflow of clinical coordinators by minimizing manual documentation and review allowing clinicians focus of care delivery.

Process:

We extracted the textual data from the heterogeneous input data and used a Gen-AI Model, to automatically generate the answers for the patient intake questionnaire. Multiple prompting techniques are implemented by our solution, to ensure the quality of our model output.

Our special retrieval augmented generation improved the performance of response generation by grounding the process in real-world data. The prompts ensure that the GPT is passed clear instructions about the format of the response. This helped the AI model generate answers that are highly accurate and relevant to the specific context of the referral.

This automation streamlines the referral review process, reduces processing time, and eliminates the need for manual input from clinical coordinators.

Result:

The implementation of Generative AI can significantly reduce processing time for referral admission decisions, enabling clinical coordinators to allocate their time and resources more efficiently. Moreover, minimizing manual documentation tasks in the overall workflow of clinical coordinators helps them to focus on delivering quality patient care.

In conclusion, our implementation of an AI-driven solution represents a paradigm shift in referral management, offering an efficient solution to address the evolving needs of healthcare organizations.